The newly-added sanitize_illegal_chars_for_xml method was being called after we

had encoded the unicode string into bytes, and doing the regular expression

substitution would then try to coerce those bytes back to unicode using the

ascii codec. Doing the sanitization on the unicode object, and then encoding to

bytes works correctly.

Ran unit tests. Before the fix, this failed with the same error as the reported

issue. After, it runs successfully.